Making and Visualizing Networks with R
PLSC 508 Political Networks
Sam Bestvater
1/27/2020
The R programming language has been growing massively in popularity as a tool for network analysis thanks to its flexibility, visualization tools, and the availability of packages like statnet developed for working with networks and relational data. In this tutorial, we will get up and running with with a basic working example exploring the core functionality of the network library. We will also explore ggnetwork, a tool for integrating network objects with the visualization library ggplot2.
Let’s start by loading some packages. statnet is a full suite of tools for working with networks. We’ll only be using functions from the network and sna libraries, but for convenience’ sake let’s go ahead and load the entire statnet suite. We’ll also load ggplot2, which is probably the most popular visualization library for R (and is one of R’s major comparative advantages against other languages such as python), and two supplemental libraries for ggplot2: GGally and ggnetwork.
# install.packages('statnet')
# install.packages('ggplot2')
# install.packages('GGally')
# install.packages('ggnetwork')
require(statnet)
require(ggplot2)
require(GGally)
require(ggnetwork)
set.seed(8675309) # set a random number seed so that visualizations appear the same each time the script runsMaking Networks in R
We’ll start by taking relational data in several different formats and using it to create a network object. We’ll be working with a network of social movement organizations active during the American Civil Rights movement, constructed from the Dynamics of Collective Action data (DCA), a protest event dataset collected by Doug McAdam, John McCarthy, Susan Olzak, and Sarah Soule (More info here). For a little background, social movement organizations (SMOs) appear as nodes in this network if they were recorded as participating in at least 2 civil rights-related protest or counterprotest by the New York Times, and organizations are linked to each other if they both appeared at the same event.
Creating a new network object
From adjacency matrix
One of the most common ways to store relational data (at least for relatively small networks) is an adjacency matrix–a square matrix where the rows and columns both represent the nodes of the network. The presence or absence of a tie between node \(i\) and node \(j\) is indicated by the value at cell \(i,j\)–in an unweighted network, such as the version of the DCA civil rights network we’re about to load, cells can take the value of either 1 or 0. Let’s go ahead and load the adjacency matrix of this network as an object in R and examine the first few rows and columns:
DCA_adj <- as.matrix(read.csv('./data/DCA_adj.csv', row.names = 1))
dim(DCA_adj)## [1] 88 88DCA_adj[1:5, 1:5]## KU.KLUX.KLAN CORE NAACP
## KU KLUX KLAN 0 1 0
## CORE 1 0 1
## NAACP 0 1 0
## AMERICANS FOR DEMOCRATIC ACTION 0 1 1
## NATIONAL STUDENT ASSOCIATION 0 1 0
## AMERICANS.FOR.DEMOCRATIC.ACTION
## KU KLUX KLAN 0
## CORE 1
## NAACP 1
## AMERICANS FOR DEMOCRATIC ACTION 0
## NATIONAL STUDENT ASSOCIATION 0
## NATIONAL.STUDENT.ASSOCIATION
## KU KLUX KLAN 0
## CORE 1
## NAACP 0
## AMERICANS FOR DEMOCRATIC ACTION 0
## NATIONAL STUDENT ASSOCIATION 0So we see that this is an 88x88 matrix representing the nodes and links in the network. Next, let’s convert this into a network object.
The core command in the network library is the network() function, which takes input data and produces a network object that can be used throughout the rest of the statnet suite. network() requires a matrix with the relational data to be specified, along with a value for matrix.type from either 'adjacency', 'edgelist', or 'incidence'. Additionally, we can supply additional vertex attributes, and boolean values for network characteristics such as whether the network is directed, bipartite, etc. This version of the DCA civil rights network is an undirected one-mode network, so our command is network(DCA_adj, matrix.type = 'adjacency', directed = F).
net1 <- network(DCA_adj, matrix.type = 'adjacency', directed = F)
net1## Network attributes:
## vertices = 88
## directed = FALSE
## hyper = FALSE
## loops = FALSE
## multiple = FALSE
## bipartite = FALSE
## total edges= 100
## missing edges= 0
## non-missing edges= 100
##
## Vertex attribute names:
## vertex.names
##
## No edge attributesIf we examine this object, we see that we have created a network object with 88 vertices and 100 edges. The vertices are named, but have no other attributes, and edges are undirected.
From edgelist
Another common format to store relational data is the edgelist, which in its simplest form is an Nx2 matrix where each row represents an edge from the node in the first column to the node in the second column. If needed, further edge attributes such as weights can be specified in additional columns. Let’s load the DCA civil rights network data in edgelist format and examine it:
DCA_el <- as.matrix(read.csv('./data/DCA_el.csv'))
dim(DCA_el)## [1] 100 2head(DCA_el)## tail_ID head_ID
## [1,] "KU KLUX KLAN" "CORE"
## [2,] "KU KLUX KLAN" "NATIONAL STATES RIGHTS PARTY"
## [3,] "CORE" "NAACP"
## [4,] "CORE" "AMERICANS FOR DEMOCRATIC ACTION"
## [5,] "CORE" "NATIONAL STUDENT ASSOCIATION"
## [6,] "CORE" "AMERICAN COMMITTEE ON AFRICA"We can see that this object contains a 100x2 matrix where each cell is the name (ID) of a vertex in the network, and each row represents an edge between two vertices.
The network() function can also take this edgelist matrix as an input, as long as we specify which columns hold the vertex IDs. The only thing we will need to change from the command we used earlier is the matrix.type argument should be set to 'edgelist'.
net2 <- network(DCA_el[,1:2], matrix.type = 'edgelist', directed = F)
net2## Network attributes:
## vertices = 61
## directed = FALSE
## hyper = FALSE
## loops = FALSE
## multiple = FALSE
## bipartite = FALSE
## total edges= 100
## missing edges= 0
## non-missing edges= 100
##
## Vertex attribute names:
## vertex.names
##
## No edge attributesWe can see that once again we have an undirected, single-mode network with 100 edges. But this time, there are only 61 vertices, rather than 88. One reason that this could happen is that edgelists, by construction, can’t contain information on nodes that are not connected to any other nodes (isolates). Since each row in the edgelist matrix represents an edge, nodes without edges simply do not appear in the edgelist representation of a network, even if they do exist in the adjacency matrix. We can check if this is what’s going on here using the isolates() command from the sna package, which takes a network object as input and returns a vector of IDs for any isolated nodes.
sna::isolates(net1)## [1] 8 13 15 16 18 21 24 25 31 35 36 38 41 49 50 59 60 67 69 72 78 80 81
## [24] 83 84 87 88It looks like several nodes in net1 are isolates.
sna::isolates(net2)## integer(0)And net2 has no isolates. Mystery solved!
Converting other network formats to network object
From pajek
As popular as R has become for network analysis, it’s certainly not the only program people use for this task, and for compatibility sometimes you’ll want to be able to load network data from other sources. One common file format for network data is the .net format from the pajeck software. Luckily, network includes a dedicated function for reading in pajek files, read.paj().
net3 <- read.paj('./data/DCA_pajek.net')
net3## Network attributes:
## vertices = 88
## directed = FALSE
## hyper = FALSE
## loops = FALSE
## multiple = FALSE
## bipartite = FALSE
## title = DCA_pajek
## total edges= 100
## missing edges= 0
## non-missing edges= 100
##
## Vertex attribute names:
## 1 2 shape vertex.names
##
## Edge attribute names:
## DCA_pajekFrom igraph
statnet is also not the only network analysis suite within R. It competes with another library called igraph. If you have data in the form of an igraph object, network unfortunately can’t read that directly. However, there is another library called intergraph which exists to translate between the formats for us. Let’s give it a try:
#install.packages('igraph')
#install.packages('intergraph')
require(igraph)
require(intergraph)In order to convert a network object into an igraph object, we use the function intergraph::asIgraph().
net4 <- asIgraph(net1)
net4## IGRAPH 1de3768 U--- 88 100 --
## + attr: na (v/l), vertex.names (v/c), na (e/l)
## + edges from 1de3768:
## [1] 1-- 2 1--55 2-- 3 2-- 4 2-- 5 2-- 7 2--12 2--14 2--17 2--20
## [11] 2--22 2--23 2--29 2--30 2--32 2--37 2--40 2--42 2--43 2--46
## [21] 2--48 2--51 2--52 2--57 2--58 2--63 2--64 2--65 2--66 2--71
## [31] 2--73 2--74 2--76 2--77 3-- 4 3-- 9 3--11 3--12 3--17 3--19
## [41] 3--22 3--23 3--27 3--29 3--30 3--39 3--44 3--46 3--48 3--51
## [51] 3--52 3--53 3--54 3--56 3--57 3--58 3--61 3--62 3--65 3--66
## [61] 3--68 3--75 3--82 4--23 5--28 6--85 9--10 12--17 12--19 12--27
## [71] 12--34 12--45 12--58 12--63 12--70 12--75 12--79 12--82 14--20 17--19
## + ... omitted several edgesWe can see that igraph objects look a little different from the network objects we’re used to at this point. Let’s convert back, using intergraph::asNetwork().
net5 <- asNetwork(net4)
net5## Network attributes:
## vertices = 88
## directed = FALSE
## hyper = FALSE
## loops = FALSE
## multiple = FALSE
## bipartite = FALSE
## total edges= 100
## missing edges= 0
## non-missing edges= 100
##
## Vertex attribute names:
## vertex.names
##
## No edge attributesThere we go, this network should be identical to the one we started with.
It’s worth noting that in addition to competing for functionality, igraph and network compete for namespace within R, meaning that there is overlap in function names between both libraries. This can create problems in your script if you try to load both packages at the same time. Generally, it’s best practice to pick one and stick with it. If you need to use both for some reason, the best way to insure you won’t run into namespace issues is by specifying the library for the commands you use with the prefixes igraph:: and network::. Right now we’ll just go ahead and remove igraph from our workspace to prevent confusion.
detach(package:igraph, unload = T)Modifying a network object
Okay, so that was a few ways to load existing network data into R as a network object. But what if we want to modify an existing network object? network makes this easy as well.
Adding new edges and vertices
Let’s say we want to add a few new nodes to the network. This is easily achieved with the add.vertices() command, which in its simplest form takes a network object and a number of new nodes as arguments:
add.vertices(net1, 2)
net1## Network attributes:
## vertices = 90
## directed = FALSE
## hyper = FALSE
## loops = FALSE
## multiple = FALSE
## bipartite = FALSE
## total edges= 100
## missing edges= 0
## non-missing edges= 100
##
## Vertex attribute names:
## vertex.names
##
## No edge attributesOkay, so now we have 90 nodes rather than 88, so that was a success, but in our DCA civil rights network, the nodes have names, and we haven’t supplied names for our new nodes yet. Let’s look at the vertex.names attribute of our network and see what that looks like now. There are a few ways to do this. First, we can use get.vertex.attribute() which takes as arguments the name of a network object and the name of an attribute we’d like to see:
get.vertex.attribute(net1, 'vertex.names')## [1] "KU.KLUX.KLAN"
## [2] "CORE"
## [3] "NAACP"
## [4] "AMERICANS.FOR.DEMOCRATIC.ACTION"
## [5] "NATIONAL.STUDENT.ASSOCIATION"
## [6] "AFL.CIO"
## [7] "AMERICAN.COMMITTEE.ON.AFRICA"
## [8] "NEW.YORK.YOUTH.COMMITTEE.FOR.INTEGRATION"
## [9] "LIQUOR.ACTION.COMMITTEE"
## [10] "NONPARTISAN.CITY.WIDE.LEADERSHIP.COMMITTEE"
## [11] "AMERICAN.JEWISH.CONGRESS"
## [12] "SCLC"
## [13] "BLACK.MUSLIMS"
## [14] "AFRICAN.FREEDOM.MOVEMENT"
## [15] "LOUISIANA.CITIZENS.COUNCIL"
## [16] "STUDENTADULT.LIASION.COMMITTEE"
## [17] "SNCC"
## [18] "COMMITTEE.ON.APPEAL.FOR.HUMAN.RIGHTS"
## [19] "CONGRESS.ON.RACIAL.EQUALITY"
## [20] "NEGRO.AMERICAN.LABOR.COUNCIL"
## [21] "COMMUNIST.PARTY"
## [22] "FREEDOM.RIDERS"
## [23] "YOUNG.PEOPLES.SOCIALIST.LEAGUE"
## [24] "PROGRESSIVE.YOUTH.ASSOCIATION"
## [25] "AMERICAN.NAZI.PARTY"
## [26] "CIVIC.INTEREST.GROUP"
## [27] "ALBANY.MOVEMENT"
## [28] "NORTHERN.STUDENT.MOVEMENT"
## [29] "UNITED.AFRICA.NATIONALIST.MOVEMENT"
## [30] "ANTIOCH.BAPTIST.CHURCH"
## [31] "EPISCOPAL.SOCIETY.FOR.CULTURAL.AND.RACIAL.UNITY"
## [32] "SHILOH.BAPTIST.CHURCH"
## [33] "MT.ZION.BAPTIST.CHURCH"
## [34] "ALABAMA.CHRISTIAN.MOVEMENT.FOR.HUMAN.RIGHTS"
## [35] "AFRICAN.NATIONALS.IN.AMERICA"
## [36] "SLCC"
## [37] "ENGLEWOOD.MOVEMENT"
## [38] "NATIONAL.RENAISSANCE.PARTY"
## [39] "URBAN.LEAGUE"
## [40] "INTERNATIONAL.UNION.OF.ELECTRICAL.RADIO.AND.MACHINE.WORKERS"
## [41] "CHATHAM.COUNTY.CRUSADE.FOR.VOTERS"
## [42] "COMMITTEE.FOR.EQUAL.EMPLOYMENT.OPPORTUNITY"
## [43] "PRESBYTERY.OF.NEW.YORK"
## [44] "NEWARK.COORDINATING.COUNCIL"
## [45] "CRUSADE.FOR.VOTERS"
## [46] "JOINT.COMMITTEE.ON.EQUAL.OPPORTUNITY"
## [47] "UNITED.FREEDOM.MOVEMENT"
## [48] "HARLEM.PARENTS.COMMITTEE"
## [49] "NONVIOLENT.ACTION.COMMITTEE"
## [50] "NATIONAL.COUNCIL.OF.CHURCHES"
## [51] "URBAN.LEAGUE.OF.NEW.YORK"
## [52] "UNITED.COMMITTEE.FOR.ACTION.NOW"
## [53] "UNITED.NEGRO.PROTEST.COMMITTEE"
## [54] "FREEDOM.NOW.COMMITTEE"
## [55] "NATIONAL.STATES.RIGHTS.PARTY"
## [56] "LONG.ISLAND.COORDINATING.COMMITTEE.FOR.CIVIL.RIGHTS"
## [57] "CITYWIDE.COMMITTEE.FOR.INTEGRATED.SCHOOLS"
## [58] "STUDENT.PEACE.UNION"
## [59] "PARENTS.AND.TAXPAYERS.MOVEMENT"
## [60] "ALLIED.ORGANIZATION.FOR.CIVIL.RIGHTS"
## [61] "FREEDOM.NOW"
## [62] "COMMITTEE.FOR.FREEDOM.NOW"
## [63] "COFO"
## [64] "PROGRESSIVE.LABOR.MOVEMENT"
## [65] "HARLEM.DEFENSE.COUNCIL"
## [66] "MISSISSIPPI.FREEDOM.DEMOCRATIC.PARTY"
## [67] "PEACE.CORPS"
## [68] "CATHOLIC.INTERRACIAL.COUNCIL"
## [69] "STUDENTS.FOR.A.DEMOCRATIC.SOCIETY"
## [70] "ST.FRANCIS.BAPTIST.CHURCH"
## [71] "FOCUS"
## [72] "YOUTH.IN.ACTION"
## [73] "BOGALUSA.VOTERS.LEAGUE"
## [74] "BOGALUSA.CIVIC.AND.VOTERS.LEAGUE"
## [75] "COORDINATING.COUNCIL.OF.COMMUNITY.ORGANIZATION"
## [76] "BOGALUSA.CIVIC.VOTERS.LEAGUE"
## [77] "COUNCIL.OF.ORGANIZATION.FOR.CIVIL.RIGHTS"
## [78] "MILWAUKEE.UNITED.SCHOOL.INTEGRATION.COMMITTEE"
## [79] "FREEDOM.MOVEMENT"
## [80] "COMMUNITY.COUNCIL.ON.HOUSING"
## [81] "BLACK.NATIONALISTS"
## [82] "CONGRESS.FOR.RACIAL.EQUALITY"
## [83] "NATIONAL.STATES.RIGHTS"
## [84] "BLACK.PANTHER.PARTY"
## [85] "ST.CATHERINES.AME.ZION.CHURCH"
## [86] "WEB.DUBOIS.CLUB"
## [87] "BROTHERS"
## [88] "ASSOCIATION.OF.BLACK.COLLEGIANS"
## [89] NA
## [90] NAThis can also be written in shorthand, making use of network’s %v% operator, and can be subset using tail() or indexing just like any other vector:
tail(net1 %v% 'vertex.names')## [1] "ST.CATHERINES.AME.ZION.CHURCH" "WEB.DUBOIS.CLUB"
## [3] "BROTHERS" "ASSOCIATION.OF.BLACK.COLLEGIANS"
## [5] NA NABoth of the above commands will work with any vertex attribute the network might have. But because names are an especially common attribute, they get their own command in addition to the general-purpose ones: network.vertex.names().
tail(network.vertex.names(net1))## [1] "ST.CATHERINES.AME.ZION.CHURCH" "WEB.DUBOIS.CLUB"
## [3] "BROTHERS" "ASSOCIATION.OF.BLACK.COLLEGIANS"
## [5] NA NAThat’s three different ways to access the same information, all of which confirm that our new nodes don’t have any names associated with them. Let’s look at how we might go about adding some:
new_names <- c('Foo', 'Bar')Let’s start by accessing the vertex.names vector again, and assigning it to a new variable:
old_names <- net1 %v% 'vertex.names'Now, lets replace the last 2 indices of this vector with our new names:
new_names <- c(old_names[1:88], new_names)
tail(new_names)## [1] "ST.CATHERINES.AME.ZION.CHURCH" "WEB.DUBOIS.CLUB"
## [3] "BROTHERS" "ASSOCIATION.OF.BLACK.COLLEGIANS"
## [5] "Foo" "Bar"Now we can re-assign the vertex.names vector with the new vector:
network.vertex.names(net1) <- new_names
tail(network.vertex.names(net1))## [1] "ST.CATHERINES.AME.ZION.CHURCH" "WEB.DUBOIS.CLUB"
## [3] "BROTHERS" "ASSOCIATION.OF.BLACK.COLLEGIANS"
## [5] "Foo" "Bar"Perfect! Now let’s add some links between these new nodes. This is done using the add.edges() command, which takes as arguments a network object, a vector of node IDs for the tails (senders) of new ties, and a vector of node IDs for the heads (receivers) of new ties. Let’s add a few:
add.edges(net1, c(90,89,90,89), c(1,5,36,28))
net1## Network attributes:
## vertices = 90
## directed = FALSE
## hyper = FALSE
## loops = FALSE
## multiple = FALSE
## bipartite = FALSE
## total edges= 104
## missing edges= 0
## non-missing edges= 104
##
## Vertex attribute names:
## vertex.names
##
## No edge attributesThere we go, now our network has 90 nodes and 104 edges.
To delete elements from a network object, we can use the commands delete.edges() and delete.vertices(). When we remove a node, all associated edges are removed along with it:
delete.vertices(net1, vid = c(89,90))
net1## Network attributes:
## vertices = 88
## directed = FALSE
## hyper = FALSE
## loops = FALSE
## multiple = FALSE
## bipartite = FALSE
## total edges= 100
## missing edges= 0
## non-missing edges= 100
##
## Vertex attribute names:
## vertex.names
##
## No edge attributesAdding other network attributes
We can also add other attributes to the nodes if we want. Let’s start by loading a dataframe containing several other characteristics for the SMOs in our network:
nodelist <- read.csv('./data/DCA_nodelist.csv', stringsAsFactors = F)
head(nodelist)## id name valence event_count deg
## 1 1 KU KLUX KLAN 2 47 4
## 2 2 CORE 1 292 66
## 3 3 NAACP 1 279 60
## 4 4 AMERICANS FOR DEMOCRATIC ACTION 1 2 6
## 5 5 NATIONAL STUDENT ASSOCIATION 1 3 4
## 6 6 AFL CIO 1 2 2Let’s start with the column called event_count. This column contains information on the number of protest events that each group is recorded as attending in the DCA data during the civil rights movement. We can see from just the first few rows that there’s a pretty wide variance here. Some central groups to the movement, such as CORE or the NAACP were involved in hundreds of protests, while others were just at a handful. Let’s add event_count as a node attribute in our network. All we need in order to do this is a vector of values for our new attribute and the set.vertex.attribute() command, which takes as arguments a network object, the name to be assigned to the new attribute, and a vector of values equal in length to the number of nodes in the network. In this case, we’ll make a new attribute called event_count and assign it the values from nodelist$event_count.
set.vertex.attribute(net1, 'event_count', nodelist$event_count)
net1## Network attributes:
## vertices = 88
## directed = FALSE
## hyper = FALSE
## loops = FALSE
## multiple = FALSE
## bipartite = FALSE
## total edges= 100
## missing edges= 0
## non-missing edges= 100
##
## Vertex attribute names:
## event_count vertex.names
##
## No edge attributesThe DCA data also contains information about the valence of each SMO toward the central issues of the protest. For example, it’s probably clear to you that while the NAACP and the Ku Klux Klan were both organizations involved in the civil rights movement, they represent very different positions on the issue. The nodelist dataframe we just loaded contains this information in the column called valence, where a value of 1 represents support for civil rights and 2 represents opposition. Let’s add this to the network as a node attribute, but first let’s recode these values as strings so they are more easily interpreted.
valence <- ifelse(nodelist$valence == 2, 'Opposition', 'Support')
set.vertex.attribute(net1, 'valence', valence)
net1## Network attributes:
## vertices = 88
## directed = FALSE
## hyper = FALSE
## loops = FALSE
## multiple = FALSE
## bipartite = FALSE
## total edges= 100
## missing edges= 0
## non-missing edges= 100
##
## Vertex attribute names:
## event_count valence vertex.names
##
## No edge attributesThis, of course, just scratches the surface of how we can modify network objects to suit our specific needs, but hopefully it provides a helpful overview of the logic underlying network objects and the tools we can apply to them.
Visualizing Networks in R
The final thing we’ll cover today is perhaps one of the most gratifying parts of beginning to work with network data: actually plotting the network and getting to visualize the connections. This is remarkably easy to do, using the plot() command built into network.
Using network::plot
At a minimum, all that network::plot() requires is a network object. It will, by default, calculate a Fruchterman-Reingold force-directed layout for the nodes, draw them, then draw the edges:
plot(net1)
We can add labels to our plot:
plot(net1,
displaylabels = T)
Okay, so that’s pretty ugly. We really don’t have enough room in this figure to include all of these names and have them be at all readable. Let’s try just labeling the groups that participated in at least 10 events. We can do this by adding a new vertex attribute to the network and using that as the labels instead:
nodelist$label <- ifelse(nodelist$event_count >= 10, nodelist$name, NA) #make a new vector of just those labels
set.vertex.attribute(net1, 'labels', nodelist$label)
plot(net1, label = net1 %v% 'labels')
That looks a little better, but some of the longer names are still getting cut off. We’ll look at one solution to this issue in a moment.
We can also change the color of vertices and edges:
plot(net1,
label = net1 %v% 'labels',
vertex.col = 'blue',
edge.col = 'grey')
You may have noticed that each time we made a change and re-drew the figure, the layout was different. This is because force-directed algorithms such as Fruchterman-Reingold have a random component and will provide a different layout solution each time they are run. If we want to be able to replicate the layout of a plot several times, we can save the coordinates and reuse them. Let’s start by creating another plot using the plot() command and assigning it to an object, xy.
xy <- plot(net1)
Now, we can make another plot use the coord argument to reuse these coordinates:
plot(net1,
label = net1 %v% 'labels',
vertex.col = 'blue',
edge.col = 'grey',
coord=xy)
There’s a lot more that can be done with network::plot() graphics–type ?plot.network.default() for more details on all the different parameters that can be adjusted.
People who use base R graphics tend to like the visualization tools included in network, since a lot of the functionality is mirrored and stuff tends to work basically the same way. I for one, am never particularly happy with the way base R plots (and, by extention network::plot() figures) end up looking. I generally think that ggplot2’s “grammar of graphics” is a better way to approach visualization, and is one of the primary comparative advantages of R as a programming language. Luckily, for folks like me, there’s ggnetwork, a function that makes it super easy to plot network objects using ggplot2. Let’s take a look:
Using ggnetwork and ggplot2
The core function of the ggnetwork library is just called ggnetwork(), and it does something pretty cool–it takes a network object as an input, runs a layout algorithm, and outputs a dataframe of x and y coordinates for the nodes and links in the network that can be directly passed to ggplot(). By default, ggnetwork() will use the Fruchterman-Reingold layout algorithm, but several others are also available. See the documentation for sna::gplot.layout() for a full list. Let’s pass our net1 object to ggnetwork() and assign the result to a new dataframe:
net1_fr <- ggnetwork(net1, layout = 'fruchtermanreingold')
head(net1_fr)## x y event_count labels na.x valence
## 1 0.6296959 0.2194328 47 KU KLUX KLAN FALSE Opposition
## 2 0.4365778 0.2735929 292 CORE FALSE Support
## 3 0.3486294 0.4635558 279 NAACP FALSE Support
## 4 0.5151933 0.4413394 2 <NA> FALSE Support
## 5 0.5684322 0.1152967 3 <NA> FALSE Support
## 6 0.3539325 0.9649099 2 <NA> FALSE Support
## vertex.names xend yend na.y
## 1 KU.KLUX.KLAN 0.6296959 0.2194328 NA
## 2 CORE 0.4365778 0.2735929 NA
## 3 NAACP 0.3486294 0.4635558 NA
## 4 AMERICANS.FOR.DEMOCRATIC.ACTION 0.5151933 0.4413394 NA
## 5 NATIONAL.STUDENT.ASSOCIATION 0.5684322 0.1152967 NA
## 6 AFL.CIO 0.3539325 0.9649099 NALet’s also try a few other layout algorithms, like the Kamada-Kawai force-directed algorithm, and just a simple circle layout:
net1_kk <- ggnetwork(net1, layout = 'kamadakawai')
head(net1_kk)## x y event_count labels na.x valence
## 1 0.2803148 0.7347220 47 KU KLUX KLAN FALSE Opposition
## 2 0.2463855 0.6045201 292 CORE FALSE Support
## 3 0.3677617 0.5270974 279 NAACP FALSE Support
## 4 0.3170791 0.5592201 2 <NA> FALSE Support
## 5 0.3229644 0.7265923 3 <NA> FALSE Support
## 6 0.9319147 0.3351788 2 <NA> FALSE Support
## vertex.names xend yend na.y
## 1 KU.KLUX.KLAN 0.2803148 0.7347220 NA
## 2 CORE 0.2463855 0.6045201 NA
## 3 NAACP 0.3677617 0.5270974 NA
## 4 AMERICANS.FOR.DEMOCRATIC.ACTION 0.3170791 0.5592201 NA
## 5 NATIONAL.STUDENT.ASSOCIATION 0.3229644 0.7265923 NA
## 6 AFL.CIO 0.9319147 0.3351788 NAnet1_cir <- ggnetwork(net1, layout = 'circle')
head(net1_cir)## x y event_count labels na.x valence
## 1 0.5000000 1.0000000 47 KU KLUX KLAN FALSE Opposition
## 2 0.5356696 0.9987261 292 CORE FALSE Support
## 3 0.5711574 0.9949107 279 NAACP FALSE Support
## 4 0.6062826 0.9885734 2 <NA> FALSE Support
## 5 0.6408663 0.9797465 3 <NA> FALSE Support
## 6 0.6747321 0.9684749 2 <NA> FALSE Support
## vertex.names xend yend na.y
## 1 KU.KLUX.KLAN 0.5000000 1.0000000 NA
## 2 CORE 0.5356696 0.9987261 NA
## 3 NAACP 0.5711574 0.9949107 NA
## 4 AMERICANS.FOR.DEMOCRATIC.ACTION 0.6062826 0.9885734 NA
## 5 NATIONAL.STUDENT.ASSOCIATION 0.6408663 0.9797465 NA
## 6 AFL.CIO 0.6747321 0.9684749 NAWe see that ggnetwork() has produced a dataframe containing coordinates for both the vertices (x and y) and edges (xend and yend), as well as all our vertex attributes. We can add these as aesthetics to a ggplot object, and the add the geom geom_nodes(), which essentially functions like geom_point().
ggplot(data = net1_fr, aes(x, y, xend=xend, yend=yend))+
geom_nodes()
This doesn’t look like too much yet, but we can see the results of the Fruchterman-Reingold layout–it’s mapping the nodes of our network into 2-dimensions, scaled between 0 and 1. The x and y axes aren’t themselves interpretable, but this is the information that ggplot (and any other graphics system we might want to use) is actually using to plot the nodes.
Like with any other geom added to a ggplot object, we can add more aesthetics to our geom_nodes(), which is really helpful for displaying node attributes in our network plot. Let’s assign valence to color and the log of event_count to size:
ggplot(data = net1_fr, aes(x, y, xend=xend, yend=yend))+
geom_nodes(aes(color = valence, size = log(event_count)))
Okay, this is showing us a good deal of information already, but it’s not a network until it has edges. These are added with the geom_edges() geom:
ggplot(data = net1_fr, aes(x, y, xend=xend, yend=yend))+
geom_nodes(aes(color = valence, size = log(event_count)))+
geom_edges()
There we go! Like before, we can make adjustments to the aesthetics of geom_edges() if we wish. If we were working with a weighted network, we might want to assign the edge weight values to size so that stronger edges appear thicker in our resulting plot. In this case, though, we don’t have any edge characteristics that vary, so we’ll just change the color. Let’s also swap the positions of geom_nodes() and geom_edges()–when ggplot() draws a figure, it draws the layer described by each geom in the order it appears in the code. Therefore, in our plot nodes are getting drawn first and then edges are getting drawn over top of them, which looks a little weird. We could probably fix this by playing around with the parameters for edge length and trying to force a gap at each node, but it’s much easier to simply have ggplot() draw the edges first, then the nodes on top. Let’s do that:
ggplot(data = net1_fr, aes(x, y, xend=xend, yend=yend))+
geom_edges(color = 'grey75')+
geom_nodes(aes(color = valence, size = log(event_count)))
Now let’s label the nodes. We can add text to our nodes with geom_nodetext(). We could supply any values we wanted here, but let’s again use the attribute we just created to label only the most active groups:
ggplot(data = net1_fr, aes(x, y, xend=xend, yend=yend))+
geom_edges(color = 'grey75')+
geom_nodes(aes(color = valence, size = log(event_count)))+
geom_nodetext(aes(label = labels))
That’s looking pretty good, but some of the names are still getting cut off. Here’s one area where ggplot2 can really shine for network visualization: it can place labels so that they “repel” each other and do not overlap or get placed outside the border of the figure. To use this feature, we use geom_nodetext_repel() instead of geom_nodetext().
ggplot(data = net1_fr, aes(x, y, xend=xend, yend=yend))+
geom_edges(color = 'grey75')+
geom_nodes(aes(color = valence, size = log(event_count)))+
geom_nodetext_repel(aes(label = labels))
That looks pretty good. Now, since it’s a standard ggplot object, we can make any other changes to it that we might wish. Let’s remove axis ticks and grid lines using theme_blank(), move the legend to the bottom of the figure using theme(legend.position='bottom'), remove the legend for centrality using guides(size=F), and add a title. Let’s also assign it a variable name so we can call it again if we want.
plot_fr <- ggplot(data = net1_fr, aes(x, y, xend=xend, yend=yend))+
geom_edges(color = 'grey75')+
geom_nodes(aes(color = valence, size = log(event_count)))+
geom_nodetext_repel(aes(label = labels))+
theme_blank()+
theme(legend.position="bottom")+
guides(size = F)+
labs(title = 'Network of co-protest ties between SMOs in the civil rights movement', color = 'Group Valence')
plot_fr
Let’s see what this looks like if we use the other layouts we calculated earlier:
plot_kk <- ggplot(data = net1_kk, aes(x, y, xend=xend, yend=yend))+
geom_edges(color = 'grey75')+
geom_nodes(aes(color = valence, size = log(event_count)))+
geom_nodetext_repel(aes(label = labels))+
theme_blank()+
theme(legend.position="bottom")+
guides(size = F)+
labs(title = 'Network of co-protest ties between SMOs in the civil rights movement', color = 'Group Valence')
plot_kk
So the Kamada-Kawai layout is similar to the Fuchterman-Reingold, but the connected component at the center of the figure seems to be layed out in a somewhat tighter configuration. How about the circle layout?
plot_cir <- ggplot(data = net1_cir, aes(x, y, xend=xend, yend=yend))+
geom_edges(color = 'grey75')+
geom_nodes(aes(color = valence, size = log(event_count)))+
geom_nodetext_repel(aes(label = labels))+
theme_blank()+
theme(legend.position="bottom")+
guides(size = F)+
labs(title = 'Network of co-protest ties between SMOs in the civil rights movement', color = 'Group Valence')
plot_cir
And finally, any ggplot object can be easily saved with ggsave(), so let’s go ahead and export the FR version of the plot as a .png:
ggsave('DCA_civilrights_network.png', plot = plot_fr, width = 10, height = 10, units = 'in', dpi = 600)